[Mplayer-cvslog] CVS: main/postproc yuv2rgb.c,1.1,1.2 yuv2rgb_template.c,1.7,1.8 Makefile,1.4,1.5
Michael Niedermayer
michael at mplayer.dev.hu
Mon Nov 26 22:17:39 CET 2001
Update of /cvsroot/mplayer/main/postproc
In directory mplayer:/var/tmp.root/cvs-serv8715
Modified Files:
yuv2rgb.c yuv2rgb_template.c Makefile
Log Message:
runtime cpu detection
Index: yuv2rgb.c
===================================================================
RCS file: /cvsroot/mplayer/main/postproc/yuv2rgb.c,v
retrieving revision 1.1
retrieving revision 1.2
diff -u -r1.1 -r1.2
--- yuv2rgb.c 6 Nov 2001 11:20:00 -0000 1.1
+++ yuv2rgb.c 26 Nov 2001 21:17:23 -0000 1.2
@@ -25,6 +25,7 @@
* along with GNU Make; see the file COPYING. If not, write to
* the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
*
+ * MMX/MMX2 Template stuff from Michael Niedermayer (michaelni at gmx.at) (needed for fast movntq support)
*/
#include <stdio.h>
@@ -34,12 +35,77 @@
#include "config.h"
//#include "video_out.h"
#include "rgb2rgb.h"
+#include "../cpudetect.h"
#ifdef HAVE_MLIB
#include "yuv2rgb_mlib.c"
#endif
-extern yuv2rgb_fun yuv2rgb_init_mmx (int bpp, int mode);
+#define DITHER1XBPP // only for mmx
+
+#ifdef ARCH_X86
+#define CAN_COMPILE_X86_ASM
+#endif
+
+#ifdef CAN_COMPILE_X86_ASM
+
+/* hope these constant values are cache line aligned */
+uint64_t __attribute__((aligned(8))) mmx_80w = 0x0080008000800080;
+uint64_t __attribute__((aligned(8))) mmx_10w = 0x1010101010101010;
+uint64_t __attribute__((aligned(8))) mmx_00ffw = 0x00ff00ff00ff00ff;
+uint64_t __attribute__((aligned(8))) mmx_Y_coeff = 0x253f253f253f253f;
+
+/* hope these constant values are cache line aligned */
+uint64_t __attribute__((aligned(8))) mmx_U_green = 0xf37df37df37df37d;
+uint64_t __attribute__((aligned(8))) mmx_U_blue = 0x4093409340934093;
+uint64_t __attribute__((aligned(8))) mmx_V_red = 0x3312331233123312;
+uint64_t __attribute__((aligned(8))) mmx_V_green = 0xe5fce5fce5fce5fc;
+
+/* hope these constant values are cache line aligned */
+uint64_t __attribute__((aligned(8))) mmx_redmask = 0xf8f8f8f8f8f8f8f8;
+uint64_t __attribute__((aligned(8))) mmx_grnmask = 0xfcfcfcfcfcfcfcfc;
+
+uint64_t __attribute__((aligned(8))) M24A= 0x00FF0000FF0000FFLL;
+uint64_t __attribute__((aligned(8))) M24B= 0xFF0000FF0000FF00LL;
+uint64_t __attribute__((aligned(8))) M24C= 0x0000FF0000FF0000LL;
+
+// the volatile is required because gcc otherwise optimizes some writes away not knowing that these
+// are read in the asm block
+volatile uint64_t __attribute__((aligned(8))) b5Dither;
+volatile uint64_t __attribute__((aligned(8))) g5Dither;
+volatile uint64_t __attribute__((aligned(8))) g6Dither;
+volatile uint64_t __attribute__((aligned(8))) r5Dither;
+
+uint64_t __attribute__((aligned(8))) dither4[2]={
+ 0x0103010301030103LL,
+ 0x0200020002000200LL,};
+
+uint64_t __attribute__((aligned(8))) dither8[2]={
+ 0x0602060206020602LL,
+ 0x0004000400040004LL,};
+
+#undef HAVE_MMX
+#undef ARCH_X86
+
+//MMX versions
+#undef RENAME
+#define HAVE_MMX
+#undef HAVE_MMX2
+#undef HAVE_3DNOW
+#define ARCH_X86
+#define RENAME(a) a ## _MMX
+#include "yuv2rgb_template.c"
+
+//MMX2 versions
+#undef RENAME
+#define HAVE_MMX
+#define HAVE_MMX2
+#undef HAVE_3DNOW
+#define ARCH_X86
+#define RENAME(a) a ## _MMX2
+#include "yuv2rgb_template.c"
+
+#endif // CAN_COMPILE_X86_ASM
uint32_t matrix_coefficients = 6;
@@ -63,10 +129,10 @@
uint8_t *, uint8_t *,
void *, void *, int);
-static void yuv2rgb_c (void * dst, uint8_t * py,
- uint8_t * pu, uint8_t * pv,
- int h_size, int v_size,
- int rgb_stride, int y_stride, int uv_stride)
+static void yuv2rgb_c (void * dst, uint8_t * py,
+ uint8_t * pu, uint8_t * pv,
+ int h_size, int v_size,
+ int rgb_stride, int y_stride, int uv_stride)
{
v_size >>= 1;
@@ -81,16 +147,29 @@
}
}
-void yuv2rgb_init (int bpp, int mode)
+void yuv2rgb_init (int bpp, int mode)
{
yuv2rgb = NULL;
-#ifdef HAVE_MMX
- if (yuv2rgb == NULL /*&& (config.flags & VO_MMX_ENABLE)*/) {
- yuv2rgb = yuv2rgb_init_mmx (bpp, mode);
- if (yuv2rgb != NULL)
- printf ("Using MMX for colorspace transform\n");
- else
- printf ("Cannot init MMX colorspace transform\n");
+#ifdef CAN_COMPILE_X86_ASM
+ if(gCpuCaps.hasMMX2)
+ {
+ if (yuv2rgb == NULL /*&& (config.flags & VO_MMX_ENABLE)*/) {
+ yuv2rgb = yuv2rgb_init_MMX2 (bpp, mode);
+ if (yuv2rgb != NULL)
+ printf ("Using MMX2 for colorspace transform\n");
+ else
+ printf ("Cannot init MMX2 colorspace transform\n");
+ }
+ }
+ else if(gCpuCaps.hasMMX)
+ {
+ if (yuv2rgb == NULL /*&& (config.flags & VO_MMX_ENABLE)*/) {
+ yuv2rgb = yuv2rgb_init_MMX (bpp, mode);
+ if (yuv2rgb != NULL)
+ printf ("Using MMX for colorspace transform\n");
+ else
+ printf ("Cannot init MMX colorspace transform\n");
+ }
}
#endif
#ifdef HAVE_MLIB
Index: yuv2rgb_template.c
===================================================================
RCS file: /cvsroot/mplayer/main/postproc/yuv2rgb_template.c,v
retrieving revision 1.7
retrieving revision 1.8
diff -u -r1.7 -r1.8
--- yuv2rgb_template.c 6 Nov 2001 19:16:49 -0000 1.7
+++ yuv2rgb_template.c 26 Nov 2001 21:17:23 -0000 1.8
@@ -24,58 +24,27 @@
* the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
*
* 15,24 bpp and dithering from Michael Niedermayer (michaelni at gmx.at)
+ * MMX/MMX2 Template stuff from Michael Niedermayer (needed for fast movntq support)
*/
-#include <stdio.h>
-#include <stdlib.h>
-
-#include "../config.h"
-
-//#include "libmpeg2/mpeg2.h"
-//#include "libmpeg2/mpeg2_internal.h"
-#include <inttypes.h>
-
-#include "rgb2rgb.h"
-#include "../mmx_defs.h"
-
-#define DITHER1XBPP
-
-/* hope these constant values are cache line aligned */
-uint64_t __attribute__((aligned(8))) mmx_80w = 0x0080008000800080;
-uint64_t __attribute__((aligned(8))) mmx_10w = 0x1010101010101010;
-uint64_t __attribute__((aligned(8))) mmx_00ffw = 0x00ff00ff00ff00ff;
-uint64_t __attribute__((aligned(8))) mmx_Y_coeff = 0x253f253f253f253f;
-
-/* hope these constant values are cache line aligned */
-uint64_t __attribute__((aligned(8))) mmx_U_green = 0xf37df37df37df37d;
-uint64_t __attribute__((aligned(8))) mmx_U_blue = 0x4093409340934093;
-uint64_t __attribute__((aligned(8))) mmx_V_red = 0x3312331233123312;
-uint64_t __attribute__((aligned(8))) mmx_V_green = 0xe5fce5fce5fce5fc;
-
-/* hope these constant values are cache line aligned */
-uint64_t __attribute__((aligned(8))) mmx_redmask = 0xf8f8f8f8f8f8f8f8;
-uint64_t __attribute__((aligned(8))) mmx_grnmask = 0xfcfcfcfcfcfcfcfc;
-
-uint64_t __attribute__((aligned(8))) M24A= 0x00FF0000FF0000FFLL;
-uint64_t __attribute__((aligned(8))) M24B= 0xFF0000FF0000FF00LL;
-uint64_t __attribute__((aligned(8))) M24C= 0x0000FF0000FF0000LL;
-
-// the volatile is required because gcc otherwise optimizes some writes away not knowing that these
-// are read in the asm block
-volatile uint64_t __attribute__((aligned(8))) b5Dither;
-volatile uint64_t __attribute__((aligned(8))) g5Dither;
-volatile uint64_t __attribute__((aligned(8))) g6Dither;
-volatile uint64_t __attribute__((aligned(8))) r5Dither;
-
-uint64_t __attribute__((aligned(8))) dither4[2]={
- 0x0103010301030103LL,
- 0x0200020002000200LL,};
-
-uint64_t __attribute__((aligned(8))) dither8[2]={
- 0x0602060206020602LL,
- 0x0004000400040004LL,};
-
+#undef MOVNTQ
+#undef EMMS
+#undef SFENCE
+
+#ifdef HAVE_3DNOW
+/* On K6 femms is faster of emms. On K7 femms is directly mapped on emms. */
+#define EMMS "femms"
+#else
+#define EMMS "emms"
+#endif
+#ifdef HAVE_MMX2
+#define MOVNTQ "movntq"
+#define SFENCE "sfence"
+#else
+#define MOVNTQ "movq"
+#define SFENCE "/nop"
+#endif
#define YUV2RGB \
/* Do the multiply part of the conversion for even and odd pixels,
@@ -152,7 +121,7 @@
"punpcklbw %%mm5, %%mm2;" /* G7 G6 G5 G4 G3 G2 G1 G0 */\
-static void yuv420_rgb16_mmx (uint8_t * image, uint8_t * py,
+static inline void RENAME(yuv420_rgb16) (uint8_t * image, uint8_t * py,
uint8_t * pu, uint8_t * pv,
int h_size, int v_size,
int rgb_stride, int y_stride, int uv_stride)
@@ -253,7 +222,7 @@
__asm__ __volatile__ (EMMS);
}
-static void yuv420_rgb15_mmx (uint8_t * image, uint8_t * py,
+static inline void RENAME(yuv420_rgb15) (uint8_t * image, uint8_t * py,
uint8_t * pu, uint8_t * pv,
int h_size, int v_size,
int rgb_stride, int y_stride, int uv_stride)
@@ -350,7 +319,7 @@
__asm__ __volatile__ (EMMS);
}
-static void yuv420_rgb24_mmx (uint8_t * image, uint8_t * py,
+static inline void RENAME(yuv420_rgb24) (uint8_t * image, uint8_t * py,
uint8_t * pu, uint8_t * pv,
int h_size, int v_size,
int rgb_stride, int y_stride, int uv_stride)
@@ -505,7 +474,7 @@
}
-static void yuv420_argb32_mmx (uint8_t * image, uint8_t * py,
+static inline void RENAME(yuv420_argb32) (uint8_t * image, uint8_t * py,
uint8_t * pu, uint8_t * pv,
int h_size, int v_size,
int rgb_stride, int y_stride, int uv_stride)
@@ -599,12 +568,12 @@
__asm__ __volatile__ (EMMS);
}
-yuv2rgb_fun yuv2rgb_init_mmx (int bpp, int mode)
+yuv2rgb_fun RENAME(yuv2rgb_init) (int bpp, int mode)
{
- if (bpp == 15 && mode == MODE_RGB) return yuv420_rgb15_mmx;
- if (bpp == 16 && mode == MODE_RGB) return yuv420_rgb16_mmx;
- if (bpp == 24 && mode == MODE_RGB) return yuv420_rgb24_mmx;
- if (bpp == 32 && mode == MODE_RGB) return yuv420_argb32_mmx;
+ if (bpp == 15 && mode == MODE_RGB) return RENAME(yuv420_rgb15);
+ if (bpp == 16 && mode == MODE_RGB) return RENAME(yuv420_rgb16);
+ if (bpp == 24 && mode == MODE_RGB) return RENAME(yuv420_rgb24);
+ if (bpp == 32 && mode == MODE_RGB) return RENAME(yuv420_argb32);
return NULL; // Fallback to C.
}
Index: Makefile
===================================================================
RCS file: /cvsroot/mplayer/main/postproc/Makefile,v
retrieving revision 1.4
retrieving revision 1.5
diff -u -r1.4 -r1.5
--- Makefile 16 Nov 2001 19:13:03 -0000 1.4
+++ Makefile 26 Nov 2001 21:17:23 -0000 1.5
@@ -6,10 +6,6 @@
SRCS=postprocess.c swscale.c rgb2rgb.c yuv2rgb.c
OBJS=$(SRCS:.c=.o)
-ifeq ($(TARGET_ARCH_X86),yes)
-SRCS += yuv2rgb_mmx.c
-endif
-
CFLAGS = $(OPTFLAGS) $(MLIB_INC) -I. -I.. -Wall $(EXTRA_INC)
# -I/usr/X11R6/include/
More information about the MPlayer-cvslog
mailing list